The Session Initiation Protocol (SIP), standardized by the IETF, allows participants to establish a session for delivering multimedia streams such as audio and video. However, when it comes to mobile environment, SIP fails to offer seamless real-time multimedia service due to significant handoff delay caused by IP address allocation and session re-establishment. A common disadvantage of the existing approaches is that they require additional network components or considerable interventions of network administrators. In this paper, we propose a fast handoff mechanism for SIP, which employs the cross-layer inter-working technique to reduce the handoff delay. The proposed mechanism, called Cross-Layer Fast Handoff for SIP (CF-SIP), proactively performs the IP address reservation and SIP session update by using link layer information of underlying wireless networks. Experimental results from our implementation show that CF-SIP can support the real-time multimedia services in wireless/mobile networks with only few modifications to existing networks. The practicability of CF-SIP is also demonstrated through its application to VoIP and H.263 video streaming.